Load Image

ComfyUI's built-in Load Image is a widget-based picker: click a thumbnail, load one file, repeat. It's fine until you need to load a directory of frames as a batch - for video work, dataset prep, or batch processing - and then it's actively in the way. Load Image from the 1hewNodes pack takes a plain path string, loads a single file or a whole folder, and hands back an image batch, a matching mask batch, and the filenames themselves.

The filename output is the sleeper feature. Stock loaders give you a filename too, but this one returns it as a string list that stays aligned with your batch - so when you save the results later you can re-attach each frame's original name. That turns out to be everything for folder round-trips.

How it works

• file - a path to an image file or a directory. Absolute paths work; relative ones resolve against ComfyUI's input directory.
• index - when loading a single image from a folder, pick by position (0-based, supports negative indices via modulo, so -1 is the last file).
• all - the big one. true loads every matched image into one batch.
• include_subdir - recurse into subfolders when file is a directory.
• get_image_size - an optional reference image. When provided, everything is center-cropped and resized to its size, so a mixed-resolution folder becomes a uniform batch. Without it, the first loaded image sets the reference size.

The mask output comes from each image's alpha/transparency (following the ComfyUI convention where transparent = white mask), plus sidecar mask files if present - files like name_mask.png or name.mask.webp next to the image are picked up and merged. Supported formats: jpg, png, webp, gif, bmp, tiff.

Outputs: image, mask, and filename (a list of filename stems).

Install

Part of ComfyUI-1hewNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/1hew/ComfyUI-1hewNodes

Restart ComfyUI. No models, no downloads.

Where it fits

The pattern it was built for: decode a video to frames → load that folder as a batch → process → save with original filenames. Or the dataset prep loop: point it at a folder, unify to a size with a reference image, batch-process, and re-save with the same names. The filename list is what makes the round-trip lossless.

Common issues

• Everything got cropped - size unification is center-crop + resize to the reference (or first) image's aspect ratio. A folder full of different aspect ratios gets its edges trimmed to match. Use fit=pad-style logic elsewhere if you must preserve full content.
• Mask is "backwards" - the alpha-derived mask is 1 - alpha, i.e. transparent pixels become white in the mask. That's the ComfyUI convention (white = mask selected = transparent), and it's the same convention the built-in loader uses. Not a bug.
• Path not found - relative paths resolve against the input directory, not the workflow folder. When in doubt, use an absolute path.

It's the rare "better Load Image" that earns its place: folder batch loading plus aligned filenames, and it doesn't try to be anything fancier.